A session at the ELLIS UnConference 2026 in Paris

AI for Tabular & Relational Data: What’s Next?

Two hours of lightning talks, open discussion, and late-breaking posters on where structured-data research goes from here.

When
Tuesday 8 December 2026Afternoon, 14:00–16:00
Where
Palais des Congrès de Paris2 Place de la Porte Maillot, 75017 Paris
Posters
In-person, late-breaking postersDue Monday 16 November 2026, AoE · 20 spots

Posters are presented in person only. We host late-breaking posters from people attending the ELLIS UnConference, so the presenting author needs to register for the UnConference and be in Paris on 8 December.

The tabular ML community as a table with one row per year. The research trend column reads boosted trees for 2016, deep learning for 2020, and foundation models for 2024; for 2027 it is a big question mark. community.csv year GPUs beats trees? hype research trend 20160—boosted trees 20201rarelydeep learning 20248oftenfoundation models 2027####NaN ?

About the session

Foundation models have transformed language and vision, and are starting to transform tabular and relational data. Where does this lead? World models over tables, unified models across modalities, or something else? This session brings the tabular and relational communities together to look back on their progress and decide what comes next.

Scope
Any table-like data, including spreadsheets, multi-table data, and relational databases.
Who should come
Researchers from the tabular and relational communities, and junior researchers from other fields looking for open problems.
You leave with
Research directions, paper ideas, and new collaborators.

Lightning talks

Six 10-minute talks that prioritize visionary thinking and critical reflection over standard technical presentations.

  • Paolo Papotti

    EURECOM

    Data management and NLP for tabular and relational data

  • Frank Hutter

    Prior Labs, ELLIS Institute Tübingen, University of Freiburg

    Tabular foundation models

  • Gaël Varoquaux

    Inria, Probabl

    Sustainability for tabular ML

  • Matthias Fey

    NVIDIA

    Graph and relational learning

  • Speaker to be announced

    Agentic data science

  • Speaker to be announced

    Multi-modal learning

Schedule

All times are Paris time (CET). Posters are presented later the same day in the UnConference poster session.

TimeSessionDetails
14:00–14:10OpeningWelcome and agenda overview
14:10–14:40Lightning talks, part 1Three 10-minute visionary research talks
14:40–14:55Group discussionPlenary Q&A and reflection
14:55–15:10BreakInformal networking
15:10–15:40Lightning talks, part 2Three 10-minute visionary research talks
15:40–16:00Group discussionPlenary Q&A and closing reflection
16:30–18:30Poster sessionIn-person late-breaking posters in the UnConference poster session

Call for in-person late-breaking posters

Coming to the ELLIS UnConference? Bring your newest work on AI for tabular and relational data.

We are looking to host late-breaking posters from people attending the UnConference in person. Late-breaking means work that has not been presented at any prior workshop or conference. This complements the UnConference’s general call for posters, which is for recently accepted papers.

Submit
A title and an abstract through the submission form.
Topics
Anything in the session’s scope, for example tabular foundation models, relational deep learning, spreadsheets and multi-table data, LLMs and agents for data science, multi-modal learning with tables, benchmarks, and efficiency.
Selection
We have space for 20 posters. If we receive more submissions, the organizers select based on scope and quality and notify everyone by 20 November.
Format
In person only, in the UnConference poster session. There is no remote option. Accepted posters are not archived and not listed online.
Attend
The presenting author registers for the ELLIS UnConference, attends in Paris, and prints their own poster.

Only for people attending the ELLIS UnConference in person. Submitting a poster here does not register you for the UnConference; register separately through its registration page.

Organizers